18F/domain-scan

A simple scanning system for the cloud

A lightweight scan pipeline for orchestrating third party tools, at scale and (optionally) using serverless infrastructure.

The point of this project is to make it easy to coordinate and parallelize third party tools with a simple scanning interface that produces consistent and database-agnostic output.

Outputs aggregate CSV for humans and machines, and detailed JSON for machines.

Can scan websites and domains for data on their HTTPS and email configuration, third party service usage, accessibility, and other things. Adding new scanners is relatively straightforward.

All scanners can be run locally using native Python multi-threading.

Some scanners can be executed inside Amazon Lambda for much higher levels of parallelization.

Most scanners work by using specialized third party tools, such as SSLyze or trustymail. Each scanner in this repo is meant to add the smallest wrapper possible around the responses returned from these tools.

There is also built-in support for using headless Chrome to efficiently measure sophisticated properties of web services. Especially powerful when combined with Amazon Lambda.

Requirements

domain-scan requires Python 3.6 or 3.7.

To install core dependencies:

pip install -r requirements.txt

You can install scanner- or gatherer-specific dependencies yourself. Or, you can "quick start" by just installing all dependencies for all scanners and/or all gatherers:

pip install -r requirements-scanners.txt
pip install -r requirements-gatherers.txt

If you plan on developing/testing domain-scan itself, install development requirements:

pip install -r requirements-dev.txt

Usage

Scan a domain. You must specify at least one "scanner" with --scan.

./scan whitehouse.gov --scan=pshtt

Scan a list of domains from a CSV. The CSV's header row will be ignored if the first cell starts with "Domain" (case-insensitive).

./scan domains.csv --scan=pshtt

Run multiple scanners on each domain:

./scan whitehouse.gov --scan=pshtt,sslyze

Append columns to each row with metadata about the scan itself, such as how long each individual scan took:

./scan example.com --scan=pshtt --meta

Scanners

• pshtt - A scanner that uses the pshtt Python package from the Department of Homeland Security's NCATS team.
• sslyze - A scanner that uses the sslyze Python package maintained by Alban Diquet.
• trustymail: The trustymail command, available from the trustymail Python package from the Department of Homeland Security's NCATS team. (Override path by setting the TRUSTYMAIL_PATH environment variable.)
• third_parties - What third party web services are in use, using headless Chrome to trap outgoing requests. (See documentation for using or writing Chrome-based scanners.)
• a11y - Accessibility issues, using pa11y.
• noop - Test scanner (no-op) used for development and debugging. Does nothing.

Parallelization

It's important to understand that scans run in parallel by default, and data is streamed to disk immediately after each scan is done.

This makes domain-scan fast, as well as memory-efficient (the entire dataset doesn't need to be read into memory), but the order of result data is unpredictable.

By default, each scanner will spin up 10 parallel threads. You can override this value with --workers. To disable this and run sequentially through each domain (1 worker), use --serial.

If row order is important to you, either disable parallelization, or use the --sort parameter to sort the resulting CSVs once the scans have completed. (Note: Using --sort will cause the entire dataset to be read into memory.)

Lambda

The domain-scan tool can execute certain compatible scanners in Amazon Lambda, instead of locally.

This can allow the use of hundreds of parallel workers, and can speed up large scans by orders of magnitude. (Assuming that the domains you're scanning are disparate enough to avoid DDoS-ing any particular service!)

See docs/lambda.md for instructions on configuring scanners for use with Amazon Lambda.

Once configured, scans can be run in Lambda using the --lambda flag, like so:

./scan example.com --scan=pshtt,sslyze --lambda

Headless Chrome

This tool has some built-in support for instrumenting headless Chrome, both locally and inside of Amazon Lambda.

Install a recent version of Node (using a user-space version manager such as nvm or nodeenv is recommended).

Then install dependencies:

npm install

Chrome-based scanners use Puppeteer, a Node-based wrapper for headless Chrome that is maintained by the Chrome team. This means that Chrome-based scanners make use of Node, even while domain-scan itself is instrumented in Python. This makes initial setup a little more complicated.

• During local scans, Python will shell out to Node from ./scanners/headless/local_bridge.py by executing ./scanners/headless/local_bridge.js, which expects the /usr/bin/env node to be usable as its executor. The data which is sent into the Node scanner, including original CLI options and environment data, is passed as a serialized JSON string as a CLI parameter, and the Node scanner returns data back to Python by emitting JSON over STDOUT.

• During Lambda scans, local execution remains exclusively in Python, and Node is never used locally. However, the Lambda function itself is expected to be in the node6.10 runtime, and uses a special Node-based Lambda handler in lambda/headless/handler.js for this purpose. There is a separate lambda/headless/deploy script for the building and deployment of Node/Chrome-based Lambda functions.

It is recommended to use Lambda in production for Chrome-based scanners -- not only for the increased speed, but because they use a simpler and cleaner method of cross-language communication (the HTTP-based function call to Amazon Lambda itself).

Support for running headless Chrome locally is intended mostly for testing and debugging with fewer moving parts (and without risk of AWS costs). Lambda support is the expected method for production scanning use cases.

See below for how to structure a Chrome-based scanner.

See docs/lambda.md for how to build and deploy Lambda-based scanners.

Options

General options:

• --scan - Required. Comma-separated names of one or more scanners.
• --sort - Sort result CSVs by domain name, alphabetically. (Note: this causes the entire dataset to be read into memory.)
• --serial - Disable parallelization, force each task to be done simultaneously. Helpful for testing and debugging.
• --debug - Print out more stuff. Useful with --serial.
• --workers - Limit parallel threads per-scanner to a number.
• --output - Where to output the cache/ and results/ directories. Defaults to ./.
• --cache - Use previously cached scan data to avoid scans hitting the network where possible.
• --suffix - Add a suffix to all input domains. For example, a --suffix of virginia.gov will add .virginia.gov to the end of all input domains.
• --lambda - Run certain scanners inside Amazon Lambda instead of locally. (See the Lambda instructions for how to use this.)
• --lambda-profile - When running Lambda-related commands, use a specified AWS named profile. Credentials/config for this named profile should already be configured separately in the execution environment.
• --meta - Append some additional columns to each row with information about the scan itself. This includes start/end times and durations, as well as any encountered errors. When also using --lambda, additional Lambda-specific information will be appended.

Output

All output files are placed into cache/ and results/ directories, whose location defaults to the current directory (./). Override the output home with --output.

• Cached full scan data about each domain is saved in the cache/ directory, named after each scan and each domain, in JSON.

Example: cache/pshtt/whitehouse.gov.json

• Formal output data in CSV form about all domains are saved in the results/ directory in CSV form, named after each scan.

Example: results/pshtt.csv

You can override the output directory by specifying --output.

It's possible for scans to save multiple CSV rows per-domain. For example, the a11y scan will have a row with details for each detected accessibility error.

• Scan metadata with the start time, end time, and scan command will be placed in the results/ directory as meta.json.

Example: results/meta.json

Using with Docker

If you're using Docker Compose, run:

docker-compose up

(You may need to use sudo.)

To scan, prefix commands with docker-compose run:

docker-compose run scan <domain> --scan=<scanner>

Gathering hostnames

This tool also includes a facility for gathering domain names that end in one or more given suffixes (e.g. .gov or yahoo.com or .gov.uk) from various sources.

By default, only fetches third-level and higher domains (excluding second-level domains).

Usage:

./gather [source] [options]

Or gather hostnames from multiple sources separated by commas:

./gather [source1,source2,...,sourceN] [options]

Right now there's one specific source (Censys.io), and then a general way of sourcing URLs or files by whatever name is convenient.

Censys.io - The censys gatherer uses data from Censys.io, which has hostnames gathered from observed certificates, through the Google BigQuery API. Censys provides certificates observed from a nightly zmap scan of the IPv4 space, as well as certificates published to public Certificate Transparency logs.

Remote or local CSV - By using any other name besides censys, this will define a gatherer based on an HTTP/HTTPS URL or local path to a CSV. Its only option is a flag named after itself. For example, using a gatherer name of dap will mean that domain-scan expects --dap to point to the URL or local file.

Hostnames found from multiple sources are deduped, and filtered by suffix or base domain according to the options given.

The resulting gathered.csv will have the following columns:

• the hostname
• the hostname's base domain
• one column for each checked source, with a value of True/False based on the hostname's presence in each source

See specific usage examples below.

General options:

• --suffix: Required. One or more suffix to filter on, separated by commas as necessary. (e.g. .gov or .yahoo.com or .gov,.gov.uk)
• --parents: A path or URL to a CSV whose first column is second-level domains. Any subdomain not contained within these second-level domains will be excluded.
• --include-parents: Include second-level domains. (Defaults to false.)
• --ignore-www: Ignore the www. prefixes of hostnames. If www.staging.example.com is found, it will be treated as staging.example.com.
• --debug: display extra output

censys: Data from Censys.io via Google BigQuery

Gathers hostnames from Censys.io via the Google BigQuery API.

Before using this, you need to:

• Create a Project in Google Cloud, and an associated service account with access to create new jobs/queries and get their results.
• Give Censys.io this Google Cloud service account to grant access to.

For details on concepts, and how to test access in the web console:

• https://support.censys.io/google-bigquery/bigquery-introduction
• https://support.censys.io/google-bigquery/adding-censys-datasets-to-bigquery

Note that the web console access is based on access given to a Google account, but BigQuery API access via this script depends on access given to Google Cloud service account credentials.

To configure access, set one of two environment variables:

• BIGQUERY_CREDENTIALS: JSON data that contains your Google BigQuery service account credentials.
• BIGQUERY_CREDENTIALS_PATH: A path to a file with JSON data that contains your Google BigQuery service account credentials.

Options:

• --timeout: Override the 10 minute job timeout (specify in seconds).
• --cache: Use locally cached data instead of hitting BigQuery.

Example:

Find hostnames ending in either .gov or .fed.us from within Censys.io's certificate database

./gather censys --suffix=.gov,.fed.us

Gathering Usage Examples

To gather .gov hostnames from Censys.io:

./gather censys --suffix=.gov --debug

To gather .gov hostnames from a hosted CSV, such as one from the Digital Analytics Program:

./gather dap --suffix=.gov --dap=https://analytics.usa.gov/data/live/sites-extended.csv

Or to gather federal-only .gov hostnames from Censys' API, a remote CSV, and a local CSV:

./gather censys,dap,private --suffix=.gov --dap=https://analytics.usa.gov/data/live/sites-extended.csv --private=/path/to/private-research.csv --parents=https://github.com/GSA/data/raw/master/dotgov-domains/current-federal.csv

a11y setup

pa11y expects a config file at config/pa11y_config.json. Details and documentation for this config can be found in the pa11y repo.

A brief note on redirects:

For the accessibility scans we're running at 18F, we're using the pshtt scanner to follow redirects before the accessibility scan runs. Pulse.cio.gov is set up to show accessibility scans for live, non-redirecting sites. For example, if aaa.gov redirects to bbb.gov, we will show results for bbb.gov on the site, but not aaa.gov.

However, if you want to include results for redirecting site, note the following. For example, if aaa.gov redirects to bbb.gov, pa11y will run against bbb.gov (but the result will be recorded for aaa.gov).

In order to get the benefits of the pshtt scanner, all a11y scans must include it. For example, to scan gsa.gov:

./scan gsa.gov --scanner=pshtt,a11y

Because of domain-scan's caching, all the results of an pshtt scan will be saved in the cache/pshtt folder, and probably does not need to be re-run for every single ally scan.

Developing new scanners

Scanners are registered by creating a single Python file in the scanners/ directory, where the file is given the name of the scanner (plus the .py extension).

(Scanners that use Chrome are slightly different, require both a Python and JavaScript file, and their differences are documented below.)

Each scanner should define a few top-level functions and one variable that will be referenced at different points.

For an example of how a scanner works, start with scanners/noop.py. The noop scanner is a test scanner that does nothing (no-op), but it implements and documents a scanner's basic Python contract.

Scanners can implement 4 functions (2 required, 2 optional). In order of being called:

• init(environment, options) (Optional)

  The init() function will be run only once, before any scans are executed.

  Returning a dict from this function will merge that dict into the environment dict passed to all subsequent function calls for every domain.

  Returning False from this function indicates that the scanner is unprepared, and the entire scan process (for all scanners) will abort.

  Useful for expensive actions that shouldn't be repeated for each scan, such as downloading supplementary data from a third party service. See the pshtt scanner for an example of downloading the Chrome preload list once, instead of for each scan.

  The init function is always run locally.

• init_domain(domain, environment, options) (Optional)

  The init_domain() function will be run once per-domain, before the scan() function is executed.

  Returning a dict from this function will merge that dict into the environment dict passed to the scan() function for that particular domain.

  Returning False from this function indicates that the domain should not be scanned. The domain will be skipped and no rows will be added to the resulting CSV. The scan function will not be called for this domain, and cached scan data for this domain will not be stored to disk.

  Useful for per-domain preparatory work that needs to be performed locally, such as taking advantage of scan information cached on disk from a prior scan. See the sslyze scanner for an example of using available pshtt data to avoid scanning a domain known not to support HTTPS.

  The init_domain function is always run locally.

• scan(domain, environment, options) (Required, unless using headless Chrome)

  The scan function performs the core of the scanning work.

  Returning a dict from this function indicates that the scan has completed successfully, and that the returned dict is the resulting information. This dict will be passed into the to_rows function described below, and used to generate one or more rows for the resulting CSV.

  Returning None from this function indicates that the scan has completed unsuccessfully. The domain will be skipped, and no rows will be added to the resulting CSV.

  In all cases, cached scan data for the domain will be stored to disk. If a scan was unsuccessful, the cached data will indicate that the scan was unsuccessful. Future scans that rely on cached responses will skip domains for which the cached scan was unsuccessful, and will not execute the scan function for those domains.

  The scan function is run either locally or in Lambda. (See docs/lambda.md for how to execute functions in Lambda.)

  If using headless Chrome, this method is defined in a corresponding Node file instead, and scan_headless must be set to True as described below.

• to_rows(data) (Required)

  The to_rows function converts the data returned by a scan into one or more rows, which will be appended to the resulting CSV.

  The data argument passed to the function is the return value of the scan function described above.

  The function must return a list of lists, where each contained list is the same length as the headers variable described below.

  For example, a to_rows function that always returns one row with two values might be as simple as return [[ data['value1'], data['value2'] ]].

  The to_rows function is always run locally.

Scanners can implement a few top-level variables (1 required, others sometimes required):

• headers (Required)

  The headers variable is a list of strings to use as column headers in the resulting CSV. These headers must be in the same order as the values in the lists returned by the to_rows function.

  The headers variable is always referenced locally.

• lambda_support (Required if using --lambda)

  Set lambda_support to True to have the scanner "opt in" to being runnable in Lambda.

  If this variable is not set, or set to False, then using --lambda will have no effect on this scanner, and it will always be run locally.

• scan_headless (Required if using headless Chrome)

  Set scan_headless to True to have the scanner indicate that its scan() method is defined in a corresponding Node file, rather than in this Python file.

  If this variable is not set, or set to False, then the scan() method must be defined. See documentation below for details on [developing Chrome scanners](#developing-chrome-scanners.

In all of the above functions that receive it, environment is a dict that will contain (at least) a scan_method key whose value is either "local" or "lambda".

The environment dict will also include any key/value pairs returned by previous function calls. This means that data returned from init will be contained in the environment dict sent to init_domain. Similarly, data returned from both init and init_domain for a particular domain will be contained in the environment dict sent to the scan method for that domain.

In all of the above functions that receive it, options is a dict that contains a direct representation of the command-line flags given to the ./scan executable.

For example, if the ./scan command is run with the flags --scan=pshtt,sslyze --lambda, they will translate to an options dict that contains (at least) {"scan": "pshtt,sslyze", "lambda": True}.

Developing Chrome scanners

This tool has some built-in support for instrumenting headless Chrome, both locally and inside of Amazon Lambda.

To make a scanner that uses headless Chrome, create two files:

• A Python file, e.g. scanners/third_parties.py, that does not have a scan() function, but does have the standard init(), init_domain(), to_rows() and headers values, as described above.

• A Node file, e.g. scanners/third_parties.js, that has a scanning function as described below.

The Node file must export the following method as part of its modules.exports:

• scan(domain, environment, options, browser, page) (Required)

  The domain, environment, and options parameters are identical to the Python equivalent. The environment dict is affected by the init() and init_domain() functions in the corresponding Python file for this scanner.

  The browser parameter is an instance of Puppeteer's Browser class. It will already be connected to a running Chromium instance.

  The page parameter is an instance of Puppeteer's Page class. It will already have been instantiated through await browser.newPage(), but not set to any particular URL.

  Returning data from this function has identical effects to its Python equivalent: the return value is sent into the to_rows() Python function, and is cached to disk as JSON, etc.

Below is a simplified example of a scan() method. A full scanner will be a bit more complicated -- see scanners/third_parties.js for a real use case.

module.exports = {
  scan: async (domain, environment, options, browser, page) => {

    // Catch each HTTP request made in the page context.
    page.on('request', (request) => {
      // process the request somehow
    });

    // Navigate to the page
    try {
      await page.goto(environment.url);
    } catch (exc) {
      // Error handling, including timeout handling.
    }

  }
}

Note that the corresponding Python file (e.g. scanners/third_parties.py) is still needed, and its init() and init_domain() functions can affect the environment object.

This can be used, for example, to provide a modified starting URL to the Node scan() function in the environment object, based on the results of previous (Python-based) scanners such as pshtt.

Public domain

This project is in the worldwide public domain. As stated in CONTRIBUTING:

This project is in the public domain within the United States, and copyright and related rights in the work worldwide are waived through the CC0 1.0 Universal public domain dedication.

All contributions to this project will be released under the CC0 dedication. By submitting a pull request, you are agreeing to comply with this waiver of copyright interest.